Tesla is reportedly working with chipmaker AMD to improve its new chips in an effort to reduce its reliance on Nvidia (Nvidia), tesla’s current GPU supplier.

Compile | AI technology base (rgznai100)


Participate in | shawn


Tesla will soon have its own self-designed task-processing chips in its cars. Tesla has received samples of the processors and is testing them, according to a person familiar with the matter. Tesla has been vertically integrating the company and reducing its reliance on other companies to develop its own chips, the sources said.

Tesla’s AI chip, however, will not be developed entirely independently, but will be built on AMD’s intellectual property. After CNBC reported the partnership, AMD shares immediately jumped, rising nearly 5 percent on the day and for several hours.

At a technology conference in Santa Clara, Calif., yesterday, Chipmaker GlobalFoundries CEO Sanjay Jha casually mentioned Tesla as a case in point. (Grofonde, a fab that AMD spun off to focus on making chips, has a wafer supply agreement with AMD through 2020.)

Grofand then had to issue a statement denying that it was commenting on customers or potential customers and that it had a direct relationship with Tesla.

It is well known that more efficient custom chips could help Tesla achieve fully autonomous driving faster. Tesla CEO Elon Musk promised earlier this year that the car would be fully autonomous by 2019.

Tesla’s recent personnel changes seem to have implications. Keller, a chip architect, became head of Tesla’s self-driving hardware and software in June, following the departure of Chris Lattner. He previously worked for Apple as the designer of the A4 and A5 chips in the iPhone.

Keller’s AI chip development team has grown to more than 50 people, including Ganesh Venkataramanan, chief hardware engineer Bill McGee, and Dan Bailey, head of system circuit design.

Neither AMD nor Tesla responded to the rumors and speculation.

Tesla, which currently uses Nvidia gpus in its cars, had Mobileye chips until tesla’s car crash last May, when it parted ways. It was acquired by Intel earlier this year for more than $15 billion. Tesla opted not to work with Intel, which has a long-standing relationship with Alphabet’s Waymo self-driving car team.

At present, Gpus made by Nvidia and other companies are widely used in a variety of AI applications. Although these chips can complete many tasks, they usually do not have the computing capability for a specific task, so the development of customized chips is imperative. If Tesla uses its own chips, it won’t be tied to Nvidia’s pricing strategy at all.


Editor: Jiao Yan